4936 words
25 minutes
Analytic Functions and Power Series Theory

Preparatory Work#

Complex Derivative#

Let UCU\subset\mathbb C, f:UCf:U\to\mathbb C, zUz\in U. If there exists a C\mathbb C-linear function dfzHomC(C,C)df|_z\in\operatorname{Hom}_{\mathbb C}(\mathbb C,\mathbb C) such that for any hCh\in\mathbb C, as h0|h|\to 0, f(z+h)f(z)dfz(h)=o(h)|f(z+h)-f(z)-df|_z(h)|=o(|h|), then ff is said to be complex differentiable at point zz, and dfz(1)df|_z(1) is called its complex derivative, denoted by f(z)f'(z). Note that the complex derivative exists if and only if there exists dfzHomC(C,C)df|_z\in\operatorname{Hom}_{\mathbb C}(\mathbb C,\mathbb C) such that

limh0f(z+h)f(z)dfz(h)h=0,\lim_{|h|\to 0}\left|\frac{f(z+h)-f(z)-df|_z(h)}{h}\right|=0,

i.e.,

limh0f(z+h)f(z)dfz(h)h=0,\lim_{h\to 0}\frac{f(z+h)-f(z)-df|_z(h)}{h}=0,

or

limh0f(z+h)f(z)h=limh0dfz(h)h.\lim_{h\to 0}\frac{f(z+h)-f(z)}{h}=\lim_{h\to 0}\frac{df|_z(h)}{h}.

Since dimCC=1\dim_{\mathbb C}\mathbb C=1, dfz(h)=f(z)hdf|_z(h)=f'(z)h, hence

f(z)=limh0f(z+h)f(z)h.f'(z)=\lim_{h\to 0}\frac{f(z+h)-f(z)}{h}.

Let f(z)=u(Rez,Imz)+iv(Rez,Imz)f(z)=u(\operatorname{Re}z,\operatorname{Im}z)+iv(\operatorname{Re}z,\operatorname{Im}z). Set z=x+iyz=x+iy, denote f~(x,y):=f(x+iy)\widetilde f(x,y):=f(x+iy). Below we will not distinguish f~\widetilde f and ff. If hh approaches zz from the real direction, then

f(z)=fx=ux+ivx,f'(z)=\frac{\partial f}{\partial x}=\frac{\partial u}{\partial x}+i\frac{\partial v}{\partial x},

if hh approaches zz from the purely imaginary direction, then

f(z)=lims0f(z+is)f(z)is=uiy+iviy=iuy+vy.f'(z)=\lim_{s\to 0}\frac{f(z+is)-f(z)}{is}=\frac{\partial u}{i\partial y}+i\frac{\partial v}{i\partial y}=-i\frac{\partial u}{\partial y}+\frac{\partial v}{\partial y}.

The equality of these two expressions gives a necessary condition for complex differentiability.

For p=x+iyCp=x+iy\in\mathbb C, formally define

fzp=12(fx(x,y)ify(x,y)),fzp=12(fx(x,y)+ify(x,y)),\frac{\partial f}{\partial z}\bigg|_{p}=\frac{1}{2}\left(\frac{\partial f}{\partial x}\bigg|_{(x,y)}-i\frac{\partial f}{\partial y}\bigg|_{(x,y)}\right),\qquad\frac{\partial f}{\partial\overline z}\bigg|_{p}=\frac{1}{2}\left(\frac{\partial f}{\partial x}\bigg|_{(x,y)}+i\frac{\partial f}{\partial y}\bigg|_{(x,y)}\right),

and define dz=dx+idy,dz=dxidydz=dx+i\,dy,\,d\overline z=dx-i\,dy, then

fzdz+fzdz=12(fxify)(dx+idy)+12(fx+ify)(dxidy)=fxdx+fydy=df.\begin{aligned} \frac{\partial f}{\partial z}\,dz+\frac{\partial f}{\partial\overline z}\,d\overline z&=\frac{1}{2}\left(\frac{\partial f}{\partial x}-i\frac{\partial f}{\partial y}\right)(dx+i\,dy)+\frac{1}{2}\left(\frac{\partial f}{\partial x}+i\frac{\partial f}{\partial y}\right)(dx-i\,dy)\\ &=\frac{\partial f}{\partial x}\,dx+\frac{\partial f}{\partial y}\,dy=df. \end{aligned}

Theorem 1.1 \quad Let UCU\subset\mathbb C be an open set, zUz\in U, then f:UCf:U\to\mathbb C is differentiable at zz if and only if f~\widetilde f is differentiable and

fz=0\frac{\partial f}{\partial\overline z}=0

i.e.,

fx=ify\frac{\partial f}{\partial x}=-i\frac{\partial f}{\partial y}

i.e.,

ux=vy,uy=vx.\frac{\partial u}{\partial x}=\frac{\partial v}{\partial y},\qquad\frac{\partial u}{\partial y}=-\frac{\partial v}{\partial x}.

Proof \quad Obviously if ff is complex differentiable at zz then f~\widetilde{f} is real differentiable at that point, hence the necessity has been shown above.

Conversely, assume f~\widetilde f is real differentiable and satisfies the Cauchy-Riemann equations. Then for hR2h\in\mathbb R^2, if h0|h|\to 0, then f~(z+h)f~=A(h)+o(h)\widetilde f(z+h)-\widetilde{f}=A(h)+o(|h|), where

A=(uxzuyzvxzvyz).A=\begin{pmatrix} \frac{\partial u}{\partial x}\big|_{z}&\frac{\partial u}{\partial y}\big|_{z}\\ \frac{\partial v}{\partial x}\big|_{z}&\frac{\partial v}{\partial y}\big|_{z} \end{pmatrix}.

Now we require A()A(\cdot) to be C\mathbb C-linear. Note that the Cauchy-Riemann equations guarantee that AA is of the form (abba)\begin{pmatrix} a&-b\\b&a \end{pmatrix}, so A()A(\cdot) is exactly (ux+ivx)()(\frac{\partial u}{\partial x}+i\frac{\partial v}{\partial x})(\cdot), thus ff is complex differentiable at zz. \square

A complex function is holomorphic if it is complex differentiable at every point of the entire UU.

Power Series#

Theorem 1.2 \quad For every power series n=0anzn\sum_{n=0}^{\infty}a_nz^n, there exists R[0,]R\in[0,\infty], called the radius of convergence, such that

  1. The series converges absolutely on B(0,R)B(0,R);
  2. The series converges uniformly on compact subsets of B(0,R)B(0,R);
  3. The series diverges on CB(0,R)\mathbb C\setminus\overline{B(0,R)}, and its terms are unbounded; d. On B(0,R)B(0,R), the sum of the series is analytic, its derivative is n=1nanzn1\sum_{n=1}^{\infty}na_nz^{n-1}, and they have the same radius of convergence;
  4. 1/R=limnsupann1/R=\lim\limits_{n\to\infty}\sup\sqrt[n]{|a_n|}.

Proof \quad If z<R|z|<R, then there exists ρ(z,R)\rho\in(|z|,R), so ρ1>R1\rho^{-1}>R^{-1}, hence NN\exists N\in\mathbb N such that nN\forall n\geq N, an1/n<ρ1|a_n|^{1/n}<\rho^{-1}, so an<ρn|a_n|<\rho^{-n}, hence anzn<(z/ρ)n|a_nz^n|<(|z|/\rho)^n. Therefore

n=Nanzn<n=N(zρ)n<,\sum_{n=N}^{\infty}\left|a_n\right||z^n|<\sum_{n=N}^{\infty}\left(\frac{|z|}{\rho}\right)^n<\infty,

so the power series converges absolutely at zz. For a given ρ<R\rho<R, choose ρ0(ρ,R)\rho_0\in(\rho,R), there exists NNN\in\mathbb N such that nN\forall n\geq N, an1/n<ρ01|a_n|^{1/n}<\rho_0^{-1}, then similarly, for any zρ|z|\leq\rho, for any ϵ>0\epsilon>0, there exists NN\mathbb N'\geq N such that n,mN\forall n,m\geq N', we have

j=nmajzj<j=nm(ρρ0)j<ϵ,\sum_{j=n}^{m}\left|a_j\right||z^j|<\sum_{j=n}^{m}\left(\frac{\rho}{\rho_0}\right)^j<\epsilon,

so the power series converges uniformly on B(0,ρ)\overline{B(0,\rho)}, hence it converges uniformly on compact subsets of B(0,R)B(0,R). If z>R|z|>R, then we can find ρ(R,z)\rho\in(R,|z|), so ρ1<R1\rho^{-1}<R^{-1}, then for any NNN\in\mathbb N there exists n>Nn>N such that an>ρn|a_n|>\rho^{-n}, so for infinitely many nn, anzn>(z/ρ)n|a_nz^n|>(|z|/\rho)^n, thus the terms of the series are unbounded, and the series diverges.

Since nn1\sqrt[n]{n}\to 1, n=1nanzn1\sum_{n=1}^{\infty}na_nz^{n-1} has the same radius of convergence. Finally, for z<R|z|<R, set

f(z)=n=0anzn=:sn(z)+Rn(z),sn(z):=a0+a1z++an1zn1,Rn(z):=j=najzj,f1(z):=n=1nanzn1=limnsn(z).\begin{aligned} f(z)&=\sum_{n=0}^{\infty}a_nz^n=:s_n(z)+R_n(z),\\ s_n(z)&:=a_0+a_1z+\cdots+a_{n-1}z^{n-1},\qquad R_n(z):=\sum_{j=n}^{\infty}a_jz^j,\\ f_1(z)&:=\sum_{n=1}^{\infty}na_nz^{n-1}=\lim_{n\to\infty}s_n'(z). \end{aligned}

Now we prove f=f1f'=f_1. For any z0B(0,R)z_0\in B(0,R), consider the identity

f(z)f(z0)zz0f1(z0)=(sn(z)sn(z0)zz0sn(z0))+(sn(z0)f1(z0))+Rn(z)Rn(z0)zz0,\begin{aligned} \frac{f(z)-f(z_0)}{z-z_0}-f_1(z_0)=\left(\frac{s_n(z)-s_n(z_0)}{z-z_0}-s_n'(z_0)\right)+(s_n'(z_0)-f_1(z_0))+\frac{R_n(z)-R_n(z_0)}{z-z_0}, \end{aligned}

where zz0z\neq z_0 and z,z0<ρ<R|z|,|z_0|<\rho<R. Note that

Rn(z)Rn(z0)zz0=1zz0j=n(ajzjajz0j)=1zz0j=n(zz0)aj(zj1+zj2z0++zz0j2+z0j1)=j=naj(zj1+zj2z0++zz0j2+z0j1),\begin{aligned} \frac{R_n(z)-R_n(z_0)}{z-z_0}&=\frac{1}{z-z_0}\sum_{j=n}^{\infty}(a_jz^j-a_jz_0^j)\\ &=\frac{1}{z-z_0}\sum_{j=n}^{\infty}(z-z_0)a_j(z^{j-1}+z^{j-2}z_0+\cdots+zz_0^{j-2}+z_0^{j-1})\\ &=\sum_{j=n}^{\infty}a_j(z^{j-1}+z^{j-2}z_0+\cdots+zz_0^{j-2}+z_0^{j-1}), \end{aligned}

thus as nn\to\infty

Rn(z)Rn(z0)zz0j=njajρj10.\left|\frac{R_n(z)-R_n(z_0)}{z-z_0}\right|\leq\sum_{j=n}^{\infty}j|a_j|\rho^{j-1}\to 0.

On the other hand, sn(z0)f1(z0)0|s_n'(z_0)-f_1(z_0)|\to 0, so there exists NNN\in\mathbb N such that for nNn\geq N, each of the above two terms is less than ϵ/3\epsilon/3. For any such nn, by the definition of derivative, there exists δ>0\delta>0 such that zB(z0,δ)\forall z\in B(z_0,\delta) we have

sn(z)sn(z0)zz0sn(z0)<ϵ3.\left|\frac{s_n(z)-s_n(z_0)}{z-z_0}-s_n'(z_0)\right|<\frac{\epsilon}{3}.

Combining these we get

f(z)f(z0)zz0f1(z0)<ϵ,\left|\frac{f(z)-f(z_0)}{z-z_0}-f_1(z_0)\right|<\epsilon,

so f1(z0)=f(z0)f_1(z_0)=f'(z_0). \square

Cauchy Integral Theorem#

Line Integrals#

Let the equation of γ:[a,b]C\gamma:[a,b]\to\mathbb C be z(t)z(t). Define

γfdz:=abf(z(t))z(t)dt\int_{\gamma}f\,dz:=\int_a^bf(z(t))z'(t)\,dt

and

γfdz:=γfdz=abf(z(t))z(t)dt.\int_{\gamma}f\,d\overline{z}:=\overline{\int_{\gamma}\overline f\,dz}=\int_a^bf(z(t))\overline{z'(t)}\,dt.

For dx=12(dz+dz),dy=12i(dzdz)dx=\frac{1}{2}(dz+d\overline{z}),\,dy=\frac{1}{2i}(dz-d\overline{z}), define

γfdx:=12(γfdz+γfdz),γfdy:=12i(γfdzγfdz),\begin{aligned} \int_{\gamma}f\,dx:=\frac{1}{2}\left(\int_{\gamma}f\,dz+\int_{\gamma}f\,d\overline{z}\right),\\ \int_{\gamma}f\,dy:=\frac{1}{2i}\left(\int_{\gamma}f\,dz-\int_{\gamma}f\,d\overline{z}\right), \end{aligned}

then

γfdz=γfdx+iγfdy,γfdz=γfdxiγfdy.\begin{aligned} \int_{\gamma}f\,dz=\int_{\gamma}f\,dx+i\int_{\gamma}f\,dy,\qquad\int_{\gamma}f\,d\overline z=\int_{\gamma}f\,dx-i\int_{\gamma}f\,dy. \end{aligned}

Set x:=Rez,y:=Imz,u:=Ref,v:=Imfx:=\operatorname{Re}z,\,y:=\operatorname{Im}z,\,u:=\operatorname{Re}f,\,v:=\operatorname{Im}f, then

fdx=abf(z(t))x(t)dt=γudx+iγvdx,fdy=abf(z(t))y(t)dt=γudy+iγvdy,fdz=γ(udxvdy)+iγ(vdx+udy),fdz=γ(udx+vdy)+iγ(vdxudy).\begin{aligned} \begin{aligned} \int f\,dx&=\int_a^bf(z(t))x'(t)\,dt=\int_{\gamma}u\,dx+i\int_{\gamma}v\,dx,\\ \int f\,dy&=\int_a^bf(z(t))y'(t)\,dt=\int_{\gamma}u\,dy+i\int_{\gamma}v\,dy,\\ \int f\,dz&=\int_{\gamma}(u\,dx-v\,dy)+i\int_{\gamma}(v\,dx+u\,dy),\\ \int f\,d\overline z&=\int_{\gamma}(u\,dx+v\,dy)+i\int_{\gamma}(v\,dx-u\,dy).\\ \end{aligned} \end{aligned}

Define the integral with respect to arc length as

γfds=γfdz:=abf(z(t))z(t)dt.\int_{\gamma}f\,ds=\int_{\gamma}f\,|dz|:=\int_a^bf(z(t))|z'(t)|\,dt.

Rectifiable Arcs#

The length of an arc is defined as

l(γ):=supn,a=t0<<tn=b1nz(tj)z(tj1).l(\gamma):=\sup_{n,a=t_0<\cdots<t_n=b}\sum_{1}^n|z(t_j)-z(t_{j-1})|.

If l(γ)<l(\gamma)<\infty, γ\gamma is said to be rectifiable.

Theorem 2.3 \quad An arc γ:z=z(t)\gamma:z=z(t) is rectifiable if and only if its real and imaginary parts are of bounded variation.

Proof \quad Note that x(tj)x(tj1)z(tj)z(tj1)|x(t_j)-x(t_{j-1})|\leq|z(t_j)-z(t_{j-1})| and y(tj)y(tj1)z(tj)z(tj1)|y(t_j)-y(t_{j-1})|\leq|z(t_j)-z(t_{j-1})|, while z(tj)z(tj1)x(tj)x(tj1)+y(tj)y(tj1)|z(t_j)-z(t_{j-1})|\leq|x(t_j)-x(t_{j-1})|+|y(t_j)-y(t_{j-1})|, thus if x(t),y(t)x(t),y(t) are of bounded variation then l(γ)<l(\gamma)<\infty, conversely, if l(γ)<l(\gamma)<\infty then x(t),y(t)x(t),y(t) are both of bounded variation. \square

Theorem 2.4 \quad If γ:[a,b]CR2\gamma:[a,b]\to\mathbb C\xrightarrow{\sim}\mathbb R^2 is C1C^1, then γ\gamma is rectifiable, and l(γ)=γds=abz(t)dtl(\gamma)=\int_{\gamma}\,ds=\int_a^b|z'(t)|\,dt.

Proof \quad Let PP be a partition of [a,b][a,b], then z(tj+1)z(tj)=tjtj+1z(t)dttjtj+1z(t)dt|z(t_{j+1})-z(t_j)|=\left|\int_{t_j}^{t_{j+1}}z'(t)\,dt\right|\leq\int_{t_j}^{t_{j+1}}|z'(t)|\,dt, so

lP(γ)=j=0n1z(tj+1)z(tj)j=0n1tjtj+1z(t)dt=abz(t)dt<,l_{P}(\gamma)=\sum_{j=0}^{n-1}|z(t_{j+1})-z(t_j)|\leq\sum_{j=0}^{n-1}\int_{t_j}^{t_{j+1}}|z'(t)|\,dt=\int_{a}^{b}|z'(t)|\,dt<\infty,

hence l(γ)<l(\gamma)<\infty, γ\gamma is rectifiable. Since z(t)z'(t) is continuous on [a,b][a,b], it is uniformly continuous, so for any ϵ>0\epsilon>0, there exists δ>0\delta>0 such that s,t[a,b],ts<δ\forall s,t\in[a,b],\,|t-s|<\delta, we have z(t)z(s)<ϵ|z'(t)-z'(s)|<\epsilon. Take a partition PP such that j,tjtj1<δ\forall j,\,|t_j-t_{j-1}|<\delta. Then for any t[tj,tj+1]t\in[t_j,t_{j+1}], z(t)z(tj)<ϵ|z'(t)-z'(t_{j})|<\epsilon. Thus

tjtj+1z(t)dttjtj+1z(tj)dt+tjtj+1z(t)z(tj)dt<tjtj+1z(tj)dt+(tj+1tj)ϵtjtj+1z(t)dt+tjtj+1(z(tj)z(t))dt+(tj+1tj)ϵ<tjtj+1z(t)dt+2(tj+1tj)ϵ=z(tj+1)z(tj)+2(tj+1tj)ϵ.\begin{aligned} \int_{t_j}^{t_{j+1}}|z'(t)|\,dt&\leq\int_{t_j}^{t_{j+1}}|z'(t_j)|\,dt+\int_{t_j}^{t_{j+1}}|z'(t)-z'(t_j)|\,dt\\ &<\left|\int_{t_j}^{t_{j+1}}z'(t_j)\,dt\right|+(t_{j+1}-t_j)\epsilon\\ &\leq\left|\int_{t_j}^{t_{j+1}}z'(t)\,dt\right|+\left|\int_{t_j}^{t_{j+1}}(z'(t_j)-z'(t))\,dt\right|+(t_{j+1}-t_j)\epsilon\\ &<\left|\int_{t_j}^{t_{j+1}}z'(t)\,dt\right|+2(t_{j+1}-t_j)\epsilon\\ &=|z(t_{j+1})-z(t_j)|+2(t_{j+1}-t_j)\epsilon. \end{aligned}

Therefore

abz(t)dt=j=0n1tjtj+1z(t)dt<j=0n1(z(tj+1)z(tj)+2(tj+1tj)ϵ)=lP(γ)+2(ba)ϵ<l(γ)+2(ba)ϵ.\begin{aligned} \int_{a}^{b}|z'(t)|\,dt&=\sum_{j=0}^{n-1}\int_{t_j}^{t_{j+1}}|z'(t)|\,dt\\ &<\sum_{j=0}^{n-1}\big(|z(t_{j+1})-z(t_j)|+2(t_{j+1}-t_j)\epsilon\big)\\ &=l_P(\gamma)+2(b-a)\epsilon<l(\gamma)+2(b-a)\epsilon. \end{aligned}

Since ϵ\epsilon is arbitrary, letting ϵ0\epsilon\to 0 yields l(γ)=γds=abz(t)dtl(\gamma)=\int_{\gamma}\,ds=\int_a^b|z'(t)|\,dt. \square

If γ\gamma is C1C^1 and f(z)f(z) is continuous on γ\gamma, then f(z(t))z(t)f(z(t))|z'(t)| is uniformly continuous.

Cauchy Integral Theorem#

After decomposing mappings into real and imaginary parts, theorems concerning exact forms, closed forms, and curve integrals can be directly applied. If p,qp,q are respectively the partial derivatives of some complex function UU with respect to xx and yy, then the integral γ(pdx+qdy)\int_{\gamma}(p\,dx+q\,dy) depends only on the endpoints. Indeed, let p=a+ib,q=c+idp=a+ib,q=c+id, then

pdx+qdy=(a+ib)dx+(c+id)dy=(adx+cdy)+i(bdx+ddy),p\,dx+q\,dy=(a+ib)\,dx+(c+id)\,dy=(a\,dx+c\,dy)+i(b\,dx+d\,dy),

if we can find V,WV,W such that dV=adx+cdy,dW=bdx+ddydV=a\,dx+c\,dy,dW=b\,dx+d\,dy, then U:=V+iWU:=V+iW satisfies dU=pdx+qdydU=p\,dx+q\,dy. Conversely, if there exists a complex-valued function UU such that dU=pdx+qdydU=p\,dx+q\,dy, then set U=V+iWU=V+iW, we have

Ux=adx+ibdx=Vx+iWx,\frac{\partial U}{\partial x}=a\,dx+ib\,dx=\frac{\partial V}{\partial x}+i\frac{\partial W}{\partial x},

giving V/x=a,W/x=b\partial V/\partial x=a,\,\partial W/\partial x=b, similarly, V/y=c,W/y=c\partial V/\partial y=c,\,\partial W/\partial y=c, so adx+cdy,bdx+ddya\,dx+c\,dy,\,b\,dx+d\,dy are each exact, hence the integral γ(pdx+qdy)\int_\gamma(p\,dx+q\,dy) is independent of the path.

If the integral γfdz\int_{\gamma}f\,dz is independent of the specific path, i.e., fdzf\,dz is exact, then there exists F(z)F(z) such that dF=fdx+ifdydF=f\,dx+if\,dy, so

Fx=f,Fy=if,\frac{\partial F}{\partial x}=f,\qquad\frac{\partial F}{\partial y}=if,

thus F(z)F(z) satisfies the Cauchy-Riemann equations.

From the above discussions, it can already be concluded that since dzdz, zdzz\,dz are exact, or more generally, zndz(n0)z^n\,dz\,(n\geq 0) are all exact, as long as the integral is closed, we have

dz=zdz=zndz=0.\oint dz=\oint z\,dz=\oint z^n\,dz=0.

Secondly, if the region is simply connected, ff is holomorphic and its partial derivatives are continuous, then fdz=0\oint f\,dz=0. This is because by the Cauchy-Riemann equations, we have respectively

d(udxvdy)=(uy+vx)dxdy=0,d(vdx+udy)=(uxvy)dxdy=0,\begin{aligned} d(u\,dx-v\,dy)&=-\left(\frac{\partial u}{\partial y}+\frac{\partial v}{\partial x}\right)dx\wedge dy=0,\\ d(v\,dx+u\,dy)&=\left(\frac{\partial u}{\partial x}-\frac{\partial v}{\partial y}\right)dx\wedge dy=0, \end{aligned}

so fdzf\,dz is exact, yielding fdz=0\oint f\,dz=0.

This is the proof initially given by Cauchy. In the corollary of Poincaré’s lemma, if UU is simply connected, then closed is equivalent to exact on UU, but the proof of Poincaré’s lemma requires the partial derivatives of ff to be continuous. Later, Goursat found that the condition of continuous partial derivatives is unnecessary, thus giving the following first version of the Cauchy integral theorem.

Theorem 2.5 (Goursat) \quad Let RR be a rectangle, ff holomorphic on R\overline{R}, then

Rfdz=0.\int_{\partial R}f\,dz=0.

Proof \quad For a region HH, denote η(H):=Hfdz\eta(H):=\int_{\partial H}f\,dz. For RR, divide it into four congruent rectangles R11,R21,R31,R41R_1^1,R_2^1,R_3^1,R_4^1, then

η(R)=η(R1)+η(R2)+η(R3)+η(R4).\eta(R)=\eta(R_1)+\eta(R_2)+\eta(R_3)+\eta(R_4).

Thus at least one Rj11R_{j_1}^1 satisfies η(Rj1)41η(R)|\eta(R_j^1)|\geq 4^{-1}|\eta(R)|. Continue dividing this Rj11R_{j_1}^1, there exists Rj22R_{j_2}^2 satisfying η(Rj22)41η(Rj22)|\eta(R_{j_2}^2)|\geq 4^{-1}|\eta(R_{j_2}^2)|. Continuing this way, we obtain a descending sequence of rectangles {Rjkk}k=1\{R_{j_k}^k\}_{k=1}^{\infty}, where η(Rjkk)4kη(R)|\eta(R_{j_k}^k)|\geq 4^{-k}|\eta(R)|. By the Cantor intersection theorem, k=1Rjkk=z0R\bigcap_{k=1}^{\infty}R_{j_k}^k=z_0\in\mathbb R. Since RR is complex differentiable at z0z_0, for any ϵ>0\epsilon>0, there exists δ>0\delta>0 such that

f(z)f(z0)f(z0)(zz0)<ϵzz0,zB(z0,δ).|f(z)-f(z_0)-f'(z_0)(z-z_0)|<\epsilon|z-z_0|,\qquad\forall z\in B(z_0,\delta).

Take nn sufficiently large such that RjnnB(z0,δ)R_{j_n}^n\subset B(z_0,\delta), let Rn:=RjnnR_n:=R_{j_n}^n. From previous discussions, we know that integrals of dzdz and zdzz\,dz are zero, so

η(Rn)=Rn(f(z)f(z0)f(z0)(zz0))dzRnf(z)f(z0)f(z0)(zz0)dz<ϵRnzz0ds.\begin{aligned} |\eta(R_n)|&=\left|\int_{\partial R_n}(f(z)-f(z_0)-f'(z_0)(z-z_0))\,dz\right|\\ &\leq\int_{\partial R_n}|f(z)-f(z_0)-f'(z_0)(z-z_0)|\,|dz|<\epsilon\int_{\partial R_n}|z-z_0|\,ds. \end{aligned}

Let the diagonal of the original rectangle be dd, perimeter be LL, then the diagonal of RnR_n is dn=2ndd_n=2^{-n}d, perimeter Ln=2nLL_n=2^{-n}L, so

4nη(R)η(Rn)<ϵRndnds=ϵdnLn=4nϵdL,4^{-n}|\eta(R)|\leq|\eta(R_n)|<\epsilon\int_{\partial R_n}d_n\,ds=\epsilon d_nL_n=4^{-n}\epsilon dL,

i.e., η(R)<ϵdL|\eta(R)|<\epsilon dL. Since ϵ\epsilon is arbitrary, we get Rfdz=η(R)=0\int_{\partial R}f\,dz=\eta(R)=0. \square

Theorem 2.6 \quad Let RR be a rectangle, with finitely many points {ζj}1n\{\zeta_j\}_1^n in its interior, ff holomorphic on R{ζj}\overline R\setminus\{\zeta_j\}. If

limzζj(zζj)f(z)=0,j=1,2,,n,\lim_{z\to\zeta_j}(z-\zeta_j)f(z)=0,\qquad\forall j=1,2,\cdots,n,

then

Rfdz=0.\int_{\partial R}f\,dz=0.

Proof \quad Decompose RR into small rectangles so that each contains at most one ζj\zeta_j. Thus the problem reduces to the case of a single point set {ζj}={ζ}\{\zeta_j\}=\{\zeta\}. In this case, divide RR into nine small rectangles, with only the central one R0R_0 containing ζ\zeta. Apply the first version of Cauchy’s integral theorem to the other eight rectangles, then sum all integrals to get

Rfdz=R0fdz.\int_{\partial R}f\,dz=\int_{\partial R_0}f\,dz.

For any ϵ\epsilon, by the theorem condition, we can choose R0R_0 sufficiently small such that on R\partial R, f(z)<ϵ/zζ|f(z)|<\epsilon/|z-\zeta|. Since such R0R_0 can be chosen arbitrarily, assume R0R_0 is a square with ζ\zeta at its center. Then

R0fdzϵdzzζ.\left|\int_{\partial R_0}f\,dz\right|\leq\epsilon\int\frac{|dz|}{|z-\zeta|}.

Now estimate the integral on the right. The integral is symmetric, so we only need to compute one side. Suppose R0R_0 has side length 2s2s, after translation ζ=0\zeta=0, consider the top side of the rectangle, then dz=dx|dz|=dx, zs|z|\geq s. So the integral on the top side is less than ssdxs=2\int_{-s}^s\frac{dx}{s}=2, overall

R0dzzζ<8,R0fdz<8ϵ.\int_{\partial R_0}\frac{|dz|}{|z-\zeta|}<8,\qquad \left|\int_{\partial R_0}f\,dz\right|<8\epsilon.

Since ϵ\epsilon is arbitrary, the proposition follows. \square

With the Cauchy integral theorem for rectangles, we can prove the Cauchy integral theorem for disks without requiring the partial derivatives of ff to be continuous.

Theorem 2.7 \quad Let ff be holomorphic on an open disk DD, then for every closed curve γ\gamma in DD, we have

γfdz=0.\int_{\gamma}f\,dz=0.

Proof \quad Let the center be x0+iy0x_0+iy_0. For x+iyDx+iy\in D, define σ\sigma as the path from x0+iy0x_0+iy_0 horizontally to x+iy0x+iy_0, then vertically to x+iyx+iy. By Goursat’s theorem, if τ\tau is the path first vertically then horizontally, then the integrals of ff along these two paths are equal. Set

F(z)=σfdz=τfdz,F(z)=\int_{\sigma}f\,dz=\int_{\tau}f\,dz,

considering the integrals along τ\tau and σ\sigma with respect to partial xx and partial yy respectively, we find

Fxz=f(z),Fyz=if(z),\frac{\partial F}{\partial x}\bigg|_{z}=f(z),\qquad\frac{\partial F}{\partial y}\bigg|_z=if(z),

so fdzf\,dz is exact, hence the integral of ff over any closed curve is zero. \square

Corollary 2.8 \quad Let DD be an open disk with finitely many points {ζj}1n\{\zeta_j\}_1^n in its interior. Let ff be holomorphic on DD, and

limzζj(zζj)f(z)=0,j=1,2,,n,\lim_{z\to\zeta_j}(z-\zeta_j)f(z)=0,\qquad\forall j=1,2,\cdots,n,

then for every closed curve γ\gamma in DD, we have

γfdz=0.\int_{\gamma}f\,dz=0.

Proof \quad Replace the Goursat theorem used in the proof of the previous theorem with its version with punctured points. \square

Winding Number#

First consider the winding number on S1S^1. In 3.1 to 3.6, we assume curves are C1C^1.

Definition 3.1 \quad Let S1R2S^1\subset\mathbb R^2 be the unit circle and c:I=[0,b]S1,t(x(t),y(t))c:I=[0,b]\to S^1,t\mapsto(x(t),y(t)) a closed curve. Find φ0\varphi_0 such that cos(φ0)=x(0),sin(φ0)=y(0)\cos(\varphi_0)=x(0),\sin(\varphi_0)=y(0). Define

φ:IR,tφ(t):=φ0+0t(xyyx)dτ,\begin{aligned} \varphi:I&\longrightarrow\mathbb R,\\ t&\longmapsto\varphi(t):=\varphi_0+\int_0^t(xy'-yx')\,d\tau, \end{aligned}

called the angle function of the closed curve cc.

The geometric meaning of the angle function is given immediately by the following proposition:

Lemma 3.2 \quad Let c:[0,b]S1c:[0,b]\to S^1 be a closed curve, φ\varphi its angle function, then

{cos(φ(t))=x(t),sin(φ(t))=y(t).\begin{cases} \cos(\varphi(t))=x(t),\\ \sin(\varphi(t))=y(t). \end{cases}

Proof \quad Set F(t):=(x(t)cos(φ(t)))2+(y(t)sin(φ(t)))2F(t):=(x(t)-\cos(\varphi(t)))^2+(y(t)-\sin(\varphi(t)))^2, then

F=(xcosφ)2+(ysinφ)2=x22xcosφ+cos2φ+y22ysinφ+sin2φ.\begin{aligned} F&=(x-\cos\varphi)^2+(y-\sin\varphi)^2\\ &=x^2-2x\cos\varphi+\cos^2\varphi+y^2-2y\sin\varphi+\sin^2\varphi. \end{aligned}

Note that φ=xyyx\varphi'=xy'-yx', differentiating FF gives

F=2xx2(xcosφxφsinφ)2φcosφsinφ+2yy2(ysinφ+yφcosφ)+2φcosφsinφ=2(xx+yy)+2(xyφ)cosφ+2(y+xφ)sinφ.\begin{aligned} F'&=2xx'-2(x'\cos\varphi-x\varphi'\sin\varphi)-2\varphi'\cos\varphi\sin\varphi\\ &\quad+2yy'-2(y'\sin\varphi+y\varphi'\cos\varphi)+2\varphi'\cos\varphi\sin\varphi\\ &=2(xx'+yy')+2(-x'-y\varphi')\cos\varphi+2(-y'+x\varphi')\sin\varphi. \end{aligned}

Since x(t)2+y(t)2=1x(t)^2+y(t)^2=1, we have 2x(t)x(t)+y(t)y(t)=02x(t)x'(t)+y(t)y'(t)=0, i.e., xx+yy=0xx'+yy'=0. x,yx,y or x,yx',y' obviously cannot be identically zero, the only possibility is that there exists k(t)k(t) such that x=ky,y=kxx'=-ky,\,y'=kx. Thus

φ(t)=xyyx=k(x2+y2)=k(t),\varphi'(t)=xy'-yx'=k(x^2+y^2)=k(t),

so x=φy,y=φxx'=-\varphi'y,\,y'=\varphi'x. Substituting yields

F0.F'\equiv 0.

This shows FF is constant. Note that at t=0t=0

F=F(0)=(x(0)cos(φ0))2+(y(0)sin(φ0))2=0,F=F(0)=(x(0)-\cos(\varphi_0))^2+(y(0)-\sin(\varphi_0))^2=0,

so FF is identically zero on [0,b][0,b]. This proves cos(φ(t))=x(t),sin(φ(t))=y(t)\cos(\varphi(t))=x(t),\sin(\varphi(t))=y(t). \square

Definition 3.3 \quad Let c:[0,b]S1c:[0,b]\to S^1 be a closed curve, φ\varphi its angle function. Define

n(c):=12π(φ(b)φ(0)),n(c):=\frac{1}{2\pi}(\varphi(b)-\varphi(0)),

called the winding number of cc. Since cc is closed, by Lemma 3.2, n(c)Zn(c)\in\mathbb Z.

Now we extend the concept of winding number to general closed curves.

Definition 3.4 \quad Let pR2p\in\mathbb R^2, γ:[0,b]R2{p}\gamma:[0,b]\to\mathbb R^2\setminus\{p\}, set d(t):=γ(t)p,ρ(t):=d(t),c(t):=d(t)/ρ(t)d(t):=\gamma(t)-p,\,\rho(t):=|d(t)|,\,c(t):=d(t)/\rho(t), then d,ρ,cC1d,\rho,c\in C^1, and γ(t)=p+ρ(t)c(t)\gamma(t)=p+\rho(t)c(t). Define

n(γ,p):=n(c)n(\gamma,p):=n(c)

as the winding number of γ\gamma about pp.

Proposition 3.5 \quad Let γ(t):=p+ρ(t)c(t)\gamma(t):=p+\rho(t)c(t) be a closed curve [0,b]R2{p}[0,b]\to\mathbb R^2\setminus\{p\}. Set (x(t),y(t)):=γ(t)p(x(t),y(t)):=\gamma(t)-p, then

n(γ,p)=12πγω0,ω0:=yx2+y2dx+xx2+y2dy.n(\gamma,p)=\frac{1}{2\pi}\int_{\gamma}\omega_0,\qquad\omega_0:=-\frac{y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy.

Proof \quad Note that c(t)=(x(t)/ρ(t),y(t)/ρ(t))c(t)=(x(t)/\rho(t),y(t)/\rho(t)), thus

n(c)=12π0b(x(t)ρ(t)(y(t)ρ(t))y(t)ρ(t)(x(t)ρ(t)))dt=12π0b(xρyρyρρ2yρxρxρρ2)dt=12π0bxyyxρ2dt.\begin{aligned} n(c)&=\frac{1}{2\pi}\int_0^b\left(\frac{x(t)}{\rho(t)}\left(\frac{y(t)}{\rho(t)}\right)'-\frac{y(t)}{\rho(t)}\left(\frac{x(t)}{\rho(t)}\right)'\right)\,dt\\ &=\frac{1}{2\pi}\int_0^b\left(\frac{x}{\rho}\frac{y'\rho-y\rho'}{\rho^2}-\frac{y}{\rho}\frac{x'\rho-x\rho'}{\rho^2}\right)\,dt=\frac{1}{2\pi}\int_0^{b}\frac{xy'-yx'}{\rho^2}\,dt. \end{aligned}

On the other hand

12πγω0=12πγ(yx2+y2dx+xx2+y2dy)=12π0b(yxx2+y2+xyx2+y2)dt=12π0bxyyxρ2dt,\begin{aligned} \frac{1}{2\pi}\int_{\gamma}\omega_0&=\frac{1}{2\pi}\int_{\gamma}\left(-\frac{y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy\right)\\ &=\frac{1}{2\pi}\int_0^b\left(-\frac{yx'}{x^2+y^2}+\frac{xy'}{x^2+y^2}\right)\,dt=\frac{1}{2\pi}\int_0^b\frac{xy'-yx'}{\rho^2}\,dt,\\ \end{aligned}

so n(γ,p)=n(c)=12πγω0n(\gamma,p)=n(c)=\frac{1}{2\pi}\int_{\gamma}\omega_0. \square

Theorem 3.6 \quad Let γ0,γ1:[0,b]R2{p}\gamma_0,\gamma_1:[0,b]\to\mathbb R^2\setminus\{p\} be two closed curves. Then γ0\gamma_0 and γ1\gamma_1 are freely homotopic if and only if n(γ0,p)=n(γ1,p)n(\gamma_0,p)=n(\gamma_1,p).

Proof \quad Take the exterior derivative of the 11-form ω0\omega_0 in Proposition 3.5:

dω0=d(yx2+y2dx+xx2+y2dy)=d(yx2+y2)dx+d(xx2+y2)dy=x2+y2(x2+y2)2dydx+x2+y2(x2+y2)2dxdy=0,\begin{aligned} d\omega_0&=d\left(-\frac{y}{x^2+y^2}\,dx+\frac{x}{x^2+y^2}\,dy\right)\\ &=d\left(\frac{-y}{x^2+y^2}\right)\wedge dx+d\left(\frac{x}{x^2+y^2}\right)\wedge dy\\ &=\frac{-x^2+y^2}{(x^2+y^2)^2}\,dy\wedge dx+\frac{-x^2+y^2}{(x^2+y^2)^2}\,dx\wedge dy=0, \end{aligned}

so ω0\omega_0 is closed. Integrals of closed forms along freely homotopic paths are always equal, so by Proposition 3.5, necessity is proven.

Sufficiency: Let γj=p+ρjcj\gamma_j=p+\rho_jc_j, let φj\varphi_j be the angle function of cjc_j, set φ(t,τ):=(1τ)φ0(t)+τφ1(t)\varphi(t,\tau):=(1-\tau)\varphi_0(t)+\tau\varphi_1(t), H(t,τ):=p+(cos(φ(t,τ)),sin(φ(t,τ)))H(t,\tau):=p+(\cos(\varphi(t,\tau)),\sin(\varphi(t,\tau))), and set γ^j:=p+cj\widehat{\gamma}_j:=p+c_j and

Hj(t,τ):=p+ρj(t)(1τ)+τρj(t)cj(t).H_j(t,\tau):=p+\frac{\rho_j(t)}{(1-\tau)+\tau\rho_j(t)}c_j(t).

Obviously HjH_j is a free homotopy from γj\gamma_j to γ^j\widehat{\gamma}_j. If n(γ0,p)=n(γ1,p)n(\gamma_0,p)=n(\gamma_1,p), then τ[0,1]\forall\tau\in[0,1] we have

φ(b,τ)φ(0,τ)=(1τ)φ0(b)+τφ1(b)((1τ)φ0(0)+τφ1(0))=(1τ)(φ0(b)φ0(0))+τ(φ1(b)φ1(0))=(1τ)2πn(γ0,p)+τ2πn(γ1,p)=2πn(γ0,p),\begin{aligned} \varphi(b,\tau)-\varphi(0,\tau)&=(1-\tau)\varphi_0(b)+\tau\varphi_1(b)-\big((1-\tau)\varphi_0(0)+\tau\varphi_1(0)\big)\\ &=(1-\tau)(\varphi_0(b)-\varphi_0(0))+\tau(\varphi_1(b)-\varphi_1(0))\\ &=(1-\tau)2\pi n(\gamma_0,p)+\tau 2\pi n(\gamma_1,p)=2\pi n(\gamma_0,p), \end{aligned}

so HH is a free homotopy from γ^0\widehat{\gamma}_0 to γ^1\widehat{\gamma}_1. In summary

γ0γ^0γ^1γ1,\gamma_0\simeq\widehat{\gamma}_0\simeq\widehat{\gamma}_1\simeq\gamma_1,

proving sufficiency. \square

Generally, let γ:[a,b]C\gamma:[a,b]\to\mathbb C be a closed rectifiable curve, wim(γ)w\notin\operatorname{im}(\gamma), define

u:[a,b]S1=DCtγ(t)γ(t)w,\begin{aligned} u:[a,b]&\longrightarrow S^1=\partial\mathbb D\subset\mathbb C\\ t&\longmapsto\frac{\gamma(t)}{|\gamma(t)-w|}, \end{aligned}

then by Theorem 3.6, if γ\gamma is once differentiable then n(γ,w)=n(u)n(\gamma,w)=n(u). Now we extend the winding number to rectifiable curves and express it using complex integrals.

Lemma 3.7 (Special case of Lifting Lemma) \quad Let π:RS1,sei2πs\pi:\mathbb R\to S^1,s\mapsto e^{i2\pi s} be the universal cover. Then there exists a lift u~:[a,b]R\tilde u:[a,b]\to\mathbb R of uu such that πu~=u\pi\circ\tilde{u}=u.

Proof \quad If uu is once differentiable then u~\tilde u can be taken as φ(t)/2π\varphi(t)/2\pi, which also proves existence for piecewise differentiable curves. Since rectifiable curves can be approximated by piecewise differentiable curves, the lemma follows. \square

Definition 3.8 \quad Let γ:[a,b]C\gamma:[a,b]\to\mathbb C, wim(γ)w\notin\operatorname{im}(\gamma), uu defined as above, u~\tilde u its lift. Define the winding number of γ\gamma about ww as n(γ,w)=u~(b)u~(a)n(\gamma,w)=\tilde u(b)-\tilde u(a), an extension of Definition 3.3.

Theorem 3.9 \quad The winding number varies continuously under free homotopy, i.e., let γj:[a,b]C\gamma_j:[a,b]\to\mathbb C, wim(γj)w\notin\operatorname{im}(\gamma_j) rectifiable, H(t,τ)H(t,\tau) a free homotopy from γ1\gamma_1 to γ2\gamma_2, then n(τ):=n(H(,τ),w)n(\tau):=n(H(\cdot,\tau),w) is continuous. In particular, if γ1,γ2\gamma_1,\gamma_2 are freely homotopic, then n(γ1,w)=n(γ2,w)n(\gamma_1,w)=n(\gamma_2,w).

Proof \quad Apply the lifting lemma to HH to get H~\tilde H, then n(τ)=H~(b,τ)H~(a,τ)n(\tau)=\tilde H(b,\tau)-\tilde H(a,\tau). Since H~\tilde H is continuous, n(τ)n(\tau) is naturally continuous. Since n(τ)n(\tau) always takes integer values, we have n(γ1,w)=n(0)=n(1)=n(γ2,w)n(\gamma_1,w)=n(0)=n(1)=n(\gamma_2,w). \square

Theorem 3.10 \quad Let γ:[a,b]C\gamma:[a,b]\to\mathbb C be a rectifiable curve, wim(γ)w\notin\operatorname{im}(\gamma). Then

n(γ,w)=12πiγ1zwdz.n(\gamma,w)=\frac{1}{2\pi i}\int_{\gamma}\frac{1}{z-w}\,dz.

Remark \quad Rectifiable curves can be uniformly approximated by piecewise differentiable curves, so it suffices to prove the proposition for the C1C^1 case. Two proofs are given here.

Proof 1 \quad Let ζ(t):=γ(t)w=x(t)+iy(t)\zeta(t):=\gamma(t)-w=x(t)+iy(t), then

dζζ=dx+idyx+iy=(dx+idy)(xiy)(x+iy)(xiy)=xdx+ydyx2+y2+i(xdyydxx2+y2)=xdx+ydyx2+y2+iω0,\begin{aligned} \frac{d\zeta}{\zeta}&=\frac{dx+i\,dy}{x+iy}=\frac{(dx+i\,dy)(x-iy)}{(x+iy)(x-iy)}\\ &=\frac{x\,dx+y\,dy}{x^2+y^2}+i\left(\frac{x\,dy-y\,dx}{x^2+y^2}\right)\\ &=\frac{x\,dx+y\,dy}{x^2+y^2}+i\omega_0, \end{aligned}

hence

12πiγ1zwdz=12πiγdζζ=12πiγiω0+12πiγxdx+ydyx2+y2=n(γ,w)+12πiγxdx+ydyx2+y2.\begin{aligned} \frac{1}{2\pi i}\int_{\gamma}\frac{1}{z-w}\,dz&=\frac{1}{2\pi i}\int_{\gamma}\frac{d\zeta}{\zeta}\\ &=\frac{1}{2\pi i}\int_{\gamma}i\omega_0+\frac{1}{2\pi i}\int_{\gamma}\frac{x\,dx+y\,dy}{x^2+y^2}\\ &=n(\gamma,w)+\frac{1}{2\pi i}\int_{\gamma}\frac{x\,dx+y\,dy}{x^2+y^2}. \end{aligned}

Note that

xdx+ydyx2+y2=12d(x2+y2)x2+y2=12dlog(x2+y2)\frac{x\,dx+y\,dy}{x^2+y^2}=\frac{1}{2}\frac{d(x^2+y^2)}{x^2+y^2}=\frac{1}{2}d\log(x^2+y^2)

is exact, so its integral around γ\gamma is zero. Therefore

n(γ,w)=12πiγ1zwdz.n(\gamma,w)=\frac{1}{2\pi i}\int_{\gamma}\frac{1}{z-w}\,dz.

Proof 2 \quad Set

h(t):=12πiatγ(s)γ(s)wds,g(t):=e2πih(t)(γ(t)w),h(t):=\frac{1}{2\pi i}\int_a^t\frac{\gamma'(s)}{\gamma(s)-w}\,ds,\qquad g(t):=e^{-2\pi ih(t)}(\gamma(t)-w),

then

h(t)=12πiγ(t)γ(t)w,g(t)0,\begin{aligned} h'(t)&=\frac{1}{2\pi i}\frac{\gamma'(t)}{\gamma(t)-w},\qquad g'(t)\equiv 0, \end{aligned}

so g(t)g(t) is constant; further g=g(a)=(γ(a)w)g=g(a)=(\gamma(a)-w). Thus

e2πih(t)=γ(t)wγ(a)w.e^{2\pi ih(t)}=\frac{\gamma(t)-w}{\gamma(a)-w}.

On the other hand

e2πiu~(t)=γ(t)wγ(t)w,e^{2\pi i\tilde u(t)}=\frac{\gamma(t)-w}{|\gamma(t)-w|},

set v(t):=logγ(t)wv(t):=\log|\gamma(t)-w|, then v(b)=v(a)v(b)=v(a) and

e2πiu~(t)+v(t)=γ(t)w,e^{2\pi i\tilde u(t)+v(t)}=\gamma(t)-w,

so there exists a constant αC\alpha\in\mathbb C such that

2πih(t)=2πiu~(t)+v(t)+α,2\pi ih(t)=2\pi i\tilde u(t)+v(t)+\alpha,

hence

n(γ,w)=u~(b)u~(a)=h(b)h(a)=12πiγ1zwdz.n(\gamma,w)=\tilde u(b)-\tilde u(a)=h(b)-h(a)=\frac{1}{2\pi i}\int_{\gamma}\frac{1}{z-w}\,dz.

\square

Cauchy Integral Formula#

This section proves that holomorphic functions are smooth, and along the way yields important theorems such as the Fundamental Theorem of Algebra.

Basic Statement#

Let f(z)f(z) be holomorphic on an open disk, γ\gamma a closed curve inside it. For a point aΔa\in\Delta outside γ\gamma, consider the function

F(z):=f(z)f(a)za,F(z):=\frac{f(z)-f(a)}{z-a},

this function is holomorphic on D{a}D\setminus\{a\}, so as zaz\to a, (za)F(z)=f(z)f(a)0(z-a)F(z)=f(z)-f(a)\to 0. Then applying the Cauchy integral theorem gives

γf(z)f(a)zadz=0,\int_{\gamma}\frac{f(z)-f(a)}{z-a}\,dz=0,

i.e.,

γf(z)zadz=f(a)γdzza.\int_{\gamma}\frac{f(z)}{z-a}\,dz=f(a)\int_{\gamma}\frac{dz}{z-a}.

Note that the integral on the right is precisely 2πin(γ,a)2\pi i\cdot n(\gamma,a). Generally, we have the following theorem.

Theorem 4.1 \quad Let ff be holomorphic on an open disk DD, γ\gamma a closed curve in DD, aCim(γ)a\in\mathbb C\setminus\operatorname{im}(\gamma), then

n(γ,a)f(a)=12πiγf(z)zadz.n(\gamma,a)f(a)=\frac{1}{2\pi i}\int_{\gamma}\frac{f(z)}{z-a}\,dz.

Proof \quad The case aDa\in D has been argued above. If aDa\notin D, then the winding number n(γ,a)n(\gamma,a) and the integral on the right are also zero, so the formula still holds. \square

For a fixed closed curve γ\gamma, for those points zz with winding number n(γ,z)=1n(\gamma,z)=1, we have

f(z)=12πiγf(w)wzdw,n(γ,z)=1.f(z)=\frac{1}{2\pi i}\int_{\gamma}\frac{f(w)}{w-z}\,dw,\qquad n(\gamma,z)=1.

This is the Cauchy integral formula. In particular, for an open disk DCD\subset\mathbb C, let f:DCf:\overline{D}\to\mathbb C be holomorphic, then taking a slightly larger open disk on which ff is also holomorphic yields

f(z)=12πiDf(w)wzdw.f(z)=\frac{1}{2\pi i}\int_{\partial D}\frac{f(w)}{w-z}\,dw.

Higher Order Derivatives#

Lemma 4.2 \quad Let ϕ(w)\phi(w) be a continuous function on an arc γ\gamma, then the function

Φn(z):=γϕ(w)(wz)ndw\Phi_n(z):=\int_{\gamma}\frac{\phi(w)}{(w-z)^n}\,dw

is analytic on Cim(γ)\mathbb C\setminus\operatorname{im}(\gamma), and Φn(z)=nΦn+1(z)\Phi_n'(z)=n\Phi_{n+1}(z).

Proof \quad First prove Φ1\Phi_1 is continuous. Let z0z_0 be a point not on γ\gamma, take B(z0,δ)B(z_0,\delta) not intersecting γ\gamma, then for all zB(z0,δ/2),wγz\in B(z_0,\delta/2),\,w\in\gamma, we have wz>δ/2|w-z|>\delta/2. Thus from

1wz1wz0=zz0(wz)(wz0)\frac{1}{w-z}-\frac{1}{w-z_0}=\frac{z-z_0}{(w-z)(w-z_0)}

we get

Φ1(z)Φ1(z0)=(zz0)γϕ(w)dw(wz)(wz0)<zz04δ2γϕ(w)dw=:Czz0,\begin{aligned} |\Phi_1(z)-\Phi_1(z_0)|&=\left|(z-z_0)\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)(w-z_0)}\right|\\ &<\left|z-z_0\right|\frac{4}{\delta^2}\int_{\gamma}|\phi(w)|\,|dw|=:C\left|z-z_0\right|, \end{aligned}

where CC is a constant. This proves Φ1\Phi_1 is continuous at any z0z_0 not on γ\gamma. Generally, from

1(wz)n1(wz0)n=1(wz)n1(wz0)+(zz0)(wz)n(wz0)1(wz0)n\frac{1}{(w-z)^n}-\frac{1}{(w-z_0)^n}=\frac{1}{(w-z)^{n-1}(w-z_0)}+\frac{(z-z_0)}{(w-z)^n(w-z_0)}-\frac{1}{(w-z_0)^n}

we get

Φn(z)Φn(z0)=[γϕ(w)dw(wz)n1(wz0)γϕ(w)dw(wz0)n]+(zz0)γϕ(w)dw(wz)n(wz0).\begin{aligned} \Phi_n(z)-\Phi_n(z_0)&=\left[\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^{n-1}(w-z_0)}-\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^n}\right]\\ &\quad+(z-z_0)\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^n(w-z_0)}. \end{aligned}

Now prove by induction that Φn\Phi_n is continuous. Assume the proposition holds for n1n-1, then taking ψ:=ϕ(w)/(wz0)\psi:=\phi(w)/(w-z_0) and applying the induction hypothesis shows the function

Ψn1(z0;z):=γϕ(w)dw(wz)n1(wz0)\Psi_{n-1}(z_0;z):=\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^{n-1}(w-z_0)}

is continuous, so as zz0z\to z_0,

γϕ(w)dw(wz)n1(wz0)γϕ(w)dw(wz0)n    γϕ(w)dw(wz0)nγϕ(w)dw(wz0)n=0,\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^{n-1}(w-z_0)}-\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^n}\;\to\;\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^n}-\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^n}=0,

and the remaining term can be shown to tend to zero as zz0z\to z_0 by taking δ/2\delta/2 similarly. This proves FnF_n is continuous.

For n=1n=1, note that

Φ1(z)Φ1(z0)zz0=γϕ(w)dw(wz)(wz0),\frac{\Phi_1(z)-\Phi_1(z_0)}{z-z_0}=\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)(w-z_0)},

we have already shown the integral on the right is continuous in zz, so as zz0z\to z_0 we get

Φ1(z0)=γϕ(w)dw(wz0)2=F2(z0),z0im(γ).\Phi_1'(z_0)=\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^2}=F_2(z_0),\qquad\forall z_0\notin\operatorname{im}(\gamma).

Now prove by induction that Φn(z)=nΦn+1(z)\Phi_n'(z)=n\Phi_{n+1}(z). Assume the proposition holds for n1n-1, then

Φn(z)Φn(z0)zz0=1zz0[γϕ(w)dw(wz)n1(wz0)γϕ(w)dw(wz0)n]+γϕ(w)dw(wz)n(wz0)=Ψn1(z0;z)Ψn1(z0;z0)zz0+γϕ(w)dw(wz)n(wz0),\begin{aligned} \frac{\Phi_n(z)-\Phi_n(z_0)}{z-z_0}&=\frac{1}{z-z_0}\left[\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^{n-1}(w-z_0)}-\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^n}\right]\\ &\quad+\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^n(w-z_0)}\\ &=\frac{\Psi_{n-1}(z_0;z)-\Psi_{n-1}(z_0;z_0)}{z-z_0}+\int_{\gamma}\frac{\phi(w)\,dw}{(w-z)^n(w-z_0)}, \end{aligned}

taking zz0z\to z_0 and using the induction hypothesis gives

Φn(z0)=(n1)Ψn(z0;z0)+Φn+1(z0).\Phi_n'(z_0)=(n-1)\Psi_{n}(z_0;z_0)+\Phi_{n+1}(z_0).

But Ψk(z0;z0)=Φk+1(z0)\Psi_k(z_0;z_0)=\Phi_{k+1}(z_0), so

Φn(z0)=nΦn+1(z0),z0im(γ).\Phi_n'(z_0)=n\Phi_{n+1}(z_0),\qquad\forall z_0\notin\operatorname{im}(\gamma).

\square

Proposition 4.3 \quad Let DCD\subset\mathbb C be an open disk, ff holomorphic on D\overline D, then for any zDz\in\mathbb D, ff is CC^{\infty} at point zz, and

f(n)(z)=n!2πiDf(ζ)dζ(ζz)n+1.f^{(n)}(z)=\frac{n!}{2\pi i}\int_{\partial D}\frac{f(\zeta)\,d\zeta}{(\zeta-z)^{n+1}}.

Proof \quad Apply the Cauchy integral theorem. \square

General Form of Cauchy Integral Theorem#

Definition 5.1 (Chain, Cycle) \quad Let UCU\subset\mathbb C be an open set, R\mathcal R the set of all rectifiable curves on UU, denote C0(U):=ZU,C1(U):=ZRC_0(U):=\mathbb Z^{\oplus U},C_1(U):=\mathbb Z^{\oplus\mathcal R}, abbreviated as C0,C1C_0,C_1. A 1-chain is a formal sum γ=j=1najγjC1\gamma=\sum_{j=1}^na_j\gamma_j\in C_1. Denote im(γ):=1nim(γj)\operatorname{im}(\gamma):=\bigcup_1^n\operatorname{im}(\gamma_j). For any fC(im(γ))f\in C(\operatorname{im}(\gamma)), define

γfdz:=j=1najγjfdz.\int_{\gamma}f\,dz:=\sum_{j=1}^na_j\int_{\gamma_j}f\,dz.

Let ι:UC0\iota:U\to C_0 be the canonical inclusion map. Define

:C1C01najγj1naj(ι(γj(βj))ι(γj(αj))),\begin{aligned} \partial:C_1&\longrightarrow C_0\\ \sum_1^na_j\gamma_j&\longmapsto\sum_1^na_j(\iota(\gamma_j(\beta_j))-\iota(\gamma_j(\alpha_j))), \end{aligned}

where γj:[αj,βj]C\gamma_j:[\alpha_j,\beta_j]\to\mathbb C. Elements in ker()\ker(\partial) are called cycles. For a cycle γ\gamma and zim(γ)z\notin\operatorname{im}(\gamma), define

n(γ,z):=12πiγdwwz.n(\gamma,z):=\frac{1}{2\pi i}\int_{\gamma}\frac{dw}{w-z}.

Definition 5.2 \quad Let UCU\subset\mathbb C be an open set, γC1(U)\gamma\in C_1(U) a cycle. γ\gamma is said to be homologous to 00 in UU if zCU\forall z\in\mathbb C\setminus U, n(γ,z)=0n(\gamma,z)=0.

In the following, we may assume the cycle γ\gamma is a sum of closed curves.

Lemma 5.3 \quad Let UCU\subset\mathbb C be a simply connected open set, then any cycle on UU is homologous to 00.

Proof \quad Each γj\gamma_j is homotopic to a point, and homotopy preserves winding number. \square

Theorem 5.4 (General Form of Cauchy Integral Theorem) \quad Let UCU\subset\mathbb C be an open set, γ\gamma a cycle on UU homologous to 00. Let ff be a holomorphic function on UU. Then

γfdz=0.\int_{\gamma}f\,dz=0.

Proof \quad First assume UU is bounded. For any δ>0\delta>0, take nZn\in\mathbb Z such that 2n<δ2^{-n}<\delta, consider the family of closed squares Qn\mathcal Q_n of side length 2n2^{-n} in the complex plane. Let {Qj}jJQn\{Q_j\}_{j\in J}\subset\mathcal Q_n be those closed squares entirely contained in UU, since UU is bounded, it is indeed a finite set. Choose δ\delta sufficiently small so that {Qj}\{Q_j\} is nonempty. Consider the cycle

Γn:=jJQj.\Gamma_{n}:=\sum_{j\in J}\partial Q_j.\\

Set

Un:=int(jJQj),Γn:=Un,Vn:=jJ(intQj).U_n:=\operatorname{int}\Bigl(\bigcup_{j\in J}Q_j\Bigr),\qquad\Gamma_n':=\partial U_n,\qquad V_n:=\bigcup_{j\in J}(\operatorname{int}Q_j).\\

Let γ\gamma be a cycle on UU homologous to 00. Choose δ\delta sufficiently small such that im(γ)Un\operatorname{im}(\gamma)\subset U_n. Let ξUUn\xi\in U\setminus U_n, then there exists QQ{Qj}Q\in\mathcal Q\setminus\{Q_j\} such that ξQ\xi\in Q and QUQ\notin U. Pick ξ0QU\xi_0\in Q\setminus U. Since γ\gamma is homologous to 00, n(γ,ξ0)=0n(\gamma,\xi_0)=0. We can connect ξ\xi and ξ0\xi_0 by a straight line LQL\subset Q such that LUn=L\cap U_{n}=\varnothing, then LCim(γ)L\subset\mathbb C\setminus\operatorname{im}(\gamma), meaning ξ0,ξ\xi_0,\xi are in the same connected component of Cim(γ)\mathbb C\setminus\operatorname{im}(\gamma). Since the winding number is continuous in the point, n(γ,ξ)=n(γ,ξ0)=0n(\gamma,\xi)=n(\gamma,\xi_0)=0. Since ΓnUUn\Gamma_n'\subset U\setminus U_n, for any ξΓn\xi\in\Gamma_n', n(γ,ξ)=0n(\gamma,\xi)=0.

Let zVnz\in V_n, then there exists a unique j0Jj_0\in J such that zQj0z\in Q_{j_0}. By the Cauchy integral formula for rectangles, we have

12πiΓnf(w)wzdw=12πiQj0f(w)wzdw=f(z).\frac{1}{2\pi i}\int_{\Gamma_{n}}\frac{f(w)}{w-z}\,dw=\frac{1}{2\pi i}\int_{\partial Q_{j_0}}\frac{f(w)}{w-z}\,dw=f(z).\\

It is easy to see that the integral along Γn\Gamma_{n} always equals the integral along Γn\Gamma_n', so

The above equality holds on VnV_n. Since both sides are continuous, it holds for all zUnz\in U_n. In particular, we have

γfdz=γ(12πiΓnf(w)wzdw)dz.\int_{\gamma}f\,dz=\int_{\gamma}\left(\frac{1}{2\pi i}\int_{\Gamma_n}\frac{f(w)}{w-z}\,dw\right)\,dz.\\

Since Γnγ=\Gamma_n'\cap\gamma=\varnothing, the integrand is continuous in both variables, so

γfdz=Γn(12πiγf(w)wzdz)dw=Γn(n(γ,w)f(w))dw=0.\begin{aligned} \int_{\gamma}f\,dz&=\int_{\Gamma_n'}\left(\frac{1}{2\pi i}\int_{\gamma}\frac{f(w)}{w-z}\,dz\right)\,dw\\ &=\int_{\Gamma_{n}'}(-n(\gamma,w)f(w))\,dw=0. \end{aligned}\\

If UU is unbounded, take RR sufficiently large such that im(γ)D(0,R)\operatorname{im}(\gamma)\subset D(0,R) and set U:=UD(0,R)U':=U\cap D(0,R). \square

Corollary 5.5 \quad Let UCU\subset\mathbb C be a simply connected open set, γ\gamma a cycle in UU, ff a holomorphic function on UU. Then

γfdz=0.\int_{\gamma}f\,dz=0.\\

Theorem 5.6 (General Form of Cauchy Integral Formula) \quad Let UCU\subset\mathbb C be an open set, γ\gamma a cycle on UU homologous to 00. Let ff be a holomorphic function on UU, aUim(γ)a\in U\setminus\operatorname{im}(\gamma). Then

12πiγf(z)zadz=n(γ,a)f(a).\frac{1}{2\pi i}\int_{\gamma}\frac{f(z)}{z-a}\,dz=n(\gamma,a)f(a).\\

Proof \quad Set

g(z):=f(z)f(a)za,g(z):=\frac{f(z)-f(a)}{z-a},\\

then gg is holomorphic on UU, so γgdz=0\int_{\gamma}g\,dz=0. Hence

12πiγf(z)zadz=12πiγf(a)dzza=n(γ,a)f(a).\frac{1}{2\pi i}\int_{\gamma}\frac{f(z)}{z-a}\,dz=\frac{1}{2\pi i}\int_{\gamma}\frac{f(a)\,dz}{z-a}=n(\gamma,a)f(a).\\

\square

Analytic Functions#

Analyticity, Taylor Expansion#

Theorem 6.1 \quad Let γ\gamma be a rectifiable curve in C\mathbb C, ϕ:im(γ)C\phi:\operatorname{im}(\gamma)\to\mathbb C a continuous function. For zCim(γ)z\in\mathbb C\setminus\operatorname{im}(\gamma), define

f(z):=γϕ(w)wzdw,f(z):=\int_{\gamma}\frac{\phi(w)}{w-z}\,dw,\\

then for any z0Cim(γ)z_0\in\mathbb C\setminus\operatorname{im}(\gamma), set r:=dist(z0,im(γ))r:=\operatorname{dist}(z_0,\operatorname{im}(\gamma)), ff can be expanded as a power series in B(z0,r)B(z_0,r):

f(z)=n=0(γϕ(w)dw(wz0)n+1)(zz0)n.f(z)=\sum_{n=0}^{\infty}\left(\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^{n+1}}\right)(z-z_0)^n.\\

Proof \quad Take wim(γ)w\in\operatorname{im}(\gamma), note that

ϕ(w)wz=ϕ(w)wz0wz0wz=ϕ(w)wz011(zz0)/(wz0),\frac{\phi(w)}{w-z}=\frac{\phi(w)}{w-z_0}\frac{w-z_0}{w-z}=\frac{\phi(w)}{w-z_0}\frac{1}{1-(z-z_0)/(w-z_0)},\\

thus for zz0wz01\left|\frac{z-z_0}{w-z_0}\right|\leq 1, in particular for zB(z0,r)z\in B(z_0,r), we have

ϕ(w)wz=n=0ϕ(w)(wz0)n+1(zz0)n.\frac{\phi(w)}{w-z}=\sum_{n=0}^{\infty}\frac{\phi(w)}{(w-z_0)^{n+1}}(z-z_0)^n.\\

Set gn(w):=j=0nϕ(w)(wz0)j1(zz0)jg_n(w):=\sum_{j=0}^{n}\phi(w)(w-z_0)^{-j-1}(z-z_0)^j, we will show gnϕ(w)/(wz)g_n\to\phi(w)/(w-z) uniformly on im(γ)\operatorname{im}(\gamma). Indeed, let M:=ϕuM:=\|\phi\|_u and r:=zz0<rr':=|z-z_0|<r, then

gn(w)ϕ(w)wz=j=nϕ(w)(wz0)j+1(zz0)jMj=n(zz0)j(wz0)j+1Mj=n(rr)j1r=Mrr(rr)n,\begin{aligned} \left|g_n(w)-\frac{\phi(w)}{w-z}\right|&=\left|\sum_{j=n}^{\infty}\frac{\phi(w)}{(w-z_0)^{j+1}}(z-z_0)^j\right|\\ &\leq M\sum_{j=n}^{\infty}\left|\frac{(z-z_0)^j}{(w-z_0)^{j+1}}\right|\leq M\sum_{j=n}^{\infty}\left(\frac{r'}{r}\right)^j\frac{1}{r}=\frac{M}{r-r'}\left(\frac{r'}{r}\right)^n, \end{aligned}\\

so gnϕ/(wz)u0\|g_n-\phi/(w-z)\|_u\to 0, proving uniform convergence. Hence

γϕ(w)wzdw=n=0(γϕ(w)dw(wz0)n+1)(zz0)n.\int_{\gamma}\frac{\phi(w)}{w-z}\,dw=\sum_{n=0}^{\infty}\left(\int_{\gamma}\frac{\phi(w)\,dw}{(w-z_0)^{n+1}}\right)(z-z_0)^n.\\

\square

Corollary 6.2 (Taylor Expansion) \quad Let D:=D(z0,r)CD:=D(z_0,r)\subset C be a disk, f:DCf:\overline D\to\mathbb C holomorphic, then ff equals a power series on DD, in particular

f(z)=12πin=0(Df(w)dw(wz0)n+1)(zz0)n.f(z)=\frac{1}{2\pi i}\sum_{n=0}^{\infty}\left(\int_{\partial D}\frac{f(w)\,dw}{(w-z_0)^{n+1}}\right)(z-z_0)^n.\\

Comparing the coefficients from Theorem 6.1 and the Cauchy integral formula with the Taylor series also yields the derivative formula

f(n)(z)=n!2πiDf(ζ)dζ(ζz)n+1,f^{(n)}(z)=\frac{n!}{2\pi i}\int_{\partial D}\frac{f(\zeta)\,d\zeta}{(\zeta-z)^{n+1}},\\

and this method is simpler than the one we used in Section 4.

Theorem 6.3 (Morera) \quad Let UCU\subset\mathbb C be an open set, f:UCf:U\to\mathbb C continuous. If for every rectifiable closed curve γ\gamma in UU, γfdz=0\int_{\gamma}f\,dz=0, then ff is holomorphic.

Proof \quad The condition implies fdzf\,dz is exact. \square

Theorem 6.4 \quad Let UCU\subset\mathbb C be an open set, {fn}\{f_n\} a sequence of holomorphic functions on UU, converging uniformly to ff on every compact subset of UU. Then ff is also a holomorphic function on UU.

Proof \quad For any DU\overline D\subset U, for zDz\in D, fn(z)=12πiDfn(w)wzdwf_n(z)=\frac{1}{2\pi i}\int_{\partial D}\frac{f_n(w)}{w-z}\,dw, letting nn\to\infty gives f(z)=12πiDf(w)wzdwf(z)=\frac{1}{2\pi i}\int_{\partial D}\frac{f(w)}{w-z}\,dw. By Theorem 6.1, ff is holomorphic on DD. Thus ff is holomorphic on the entire UU. \square

Cauchy Estimates, Liouville’s Theorem, Fundamental Theorem of Algebra#

Theorem 6.5 (Cauchy’s estimate) \quad Let D:=B(z0,r)CD:=B(z_0,r)\subset\mathbb C be a disk, f:DCf:\overline{D}\to\mathbb C holomorphic. Set M:=maxzDf(z)M:=\max_{z\in\partial D}|f(z)|. Then

f(n)(z0)Mn!rn.|f^{(n)}(z_0)|\leq Mn!r^{-n}.\\

Proof \quad

f(n)(z0)=n!2πiDf(ζ)dζ(ζz)n+1DMrn1ds=Mn!rn.\begin{aligned} |f^{(n)}(z_0)|&=\left|\frac{n!}{2\pi i}\int_{\partial D}\frac{f(\zeta)\,d\zeta}{(\zeta-z)^{n+1}}\right|\leq\left|\int_{\partial D}Mr^{-n-1}\,ds\right|=Mn!r^{-n}. \end{aligned}\\

\square

Theorem 6.6 (Liouville) \quad Let f:CCf:\mathbb C\to\mathbb C be a bounded holomorphic function, then ff is constant.

Proof \quad Suppose fM|f|\leq M, let z0Cz_0\in\mathbb C, for any RRR\in\mathbb R apply Cauchy’s estimate on B(z0,R)B(z_0,R) to get f(z0)MRn|f'(z_0)|\leq MR^{-n}, since RR is arbitrary, f(z0)=0f'(z_0)=0, so f0f'\equiv 0. Hence ff is constant. \square

Definition 6.7 \quad If a function ff is holomorphic on the entire complex plane, it is called an entire function.

Corollary 6.8 \quad If the real or imaginary part of an entire function ff is bounded, then ff is constant.

Proof \quad Let f=u+ivf=u+iv. Suppose uu is bounded. Consider g:=exp(f)g:=\exp(f), then g=eu|g|=|e^u|. Since uu is bounded, so is gg, thus by Liouville’s theorem g=exp(f)g=\exp(f) is constant, so ff is constant. \square

Theorem 6.9 (Fundamental Theorem of Algebra) \quad C\mathbb C is algebraically closed: Let fC[X]f\in\mathbb C[X], degf1\deg f\geq 1. Then ff has at least one root in C\mathbb C.

Proof \quad Assume ff has no root in C\mathbb C, then 1/f1/f is holomorphic on all of C\mathbb C. Since deg1\deg\geq 1, as z|z|\to\infty, f|f|\to\infty, so 1/f01/f\to 0, hence 1/f1/f is bounded. By Liouville’s theorem, 1/f1/f is constant, so ff is also constant, contradicting degf1\deg f\geq 1. Therefore ff has at least one root in C\mathbb C. \square

Zeros and Uniqueness#

Lemma 6.10 \quad Let UCU\subset\mathbb C be a connected open set, ff a holomorphic function on UU, z0Uz_0\in U. Suppose for all n1n\geq 1, f(n)(z0)=0f^{(n)}(z_0)=0. Then ff(z0)f\equiv f(z_0) on UU.

Proof \quad Set

V:={zU:f(z)=f(z0),f(n)(z)=0,n1}.V:=\{z\in U:f(z)=f(z_0),\,f^{(n)}(z)=0,\,\forall n\geq 1\}.\\

Since z0Vz_0\in V, VV is nonempty. Clearly VV is closed. Let zVz\in V, then there exists D(z,ϵ)U\overline{D(z,\epsilon)}\subset U such that ff is holomorphic on D(z,ϵ)\overline{D(z,\epsilon)}, so ff can be expanded as a power series in DD and all coefficients of nonzero terms are zero, hence ff is identically f(z0)f(z_0) on DD, so all derivatives of ff are zero on DD, thus DVD\subset V. This shows VV is open. Since UU is connected, V=UV=U. \square

Lemma 6.11 \quad Zeros of holomorphic functions are isolated: Let UCU\subset\mathbb C be a connected open set, ff a non-constant holomorphic function on UU, z0Uz_0\in U, then there exists r>0r>0 such that z0z_0 is the unique zero of ff(z0)f-f(z_0) in D(z0,r)D(z_0,r).

Proof \quad By Lemma 6.10, there exist r0>0r_0>0 and mN1m\in\mathbb N_{\geq 1} such that on D:=D(z0,r0)D':=D(z_0,r_0)

f(z)=f(z0)+am(zz0)m+,am0,f(z)=f(z_0)+a_m(z-z_0)^m+\cdots,\quad a_m\neq 0,\\

then on DD', ff(z0)=(zz0)mg(z)f-f(z_0)=(z-z_0)^mg(z), where the holomorphic function gg satisfies g(z0)=am0g(z_0)=a_m\geq 0. Since gg is continuous, there exists 0<r<r00<r<r_0 such that on D:=D(z0,r)D:=D(z_0,r), g(z)0g(z)\neq 0. Thus z0z_0 is the unique zero of ff(z0)f-f(z_0) on DD. \square

Corollary 6.12 \quad Let UCU\subset\mathbb C be a connected open set, f,gf,g holomorphic functions on UU. If there exists a sequence of points {zn}\{z_n\} such that znzUz_n\to z\in U and f(zj)=g(zj),jNf(z_j)=g(z_j),\quad\forall j\in\mathbb N, then fgf\equiv g on UU.

Proof \quad By continuity f(z0)=g(z0)f(z_0)=g(z_0), then apply Lemma 6.11. \square

Logarithm and Taking Roots#

Theorem 6.13 (Logarithm) \quad Let UCU\subset\mathbb C be a simply connected open set, ff a holomorphic function on UU, with f0f\neq 0 pointwise. Then there exists a holomorphic function gg on UU such that eg=fe^g=f.

Proof \quad f/ff'/f is holomorphic on UU. For aUa\in U, choose αC\alpha\in\mathbb C such that eα=f(a)e^{\alpha}=f(a). For any zUz\in U, let γ\gamma be a rectifiable curve connecting aa and zz, define

g(z):=α+γffdz,g(z):=\alpha+\int_{\gamma}\frac{f'}{f}\,dz,\\

then gg is independent of the choice of γ\gamma, so gg is holomorphic and g=f/fg'=f'/f. Set h:=exp(g)fh:=\exp(-g)f, then h(a)=1h(a)=1. Further

h(z)=eg(z)g(z)f(z)+f(z)eg(z)=0,zU.h'(z)=-e^{-g(z)}g'(z)f(z)+f'(z)e^{-g(z)}=0,\qquad\forall z\in U.\\

Thus h1h\equiv 1. Therefore eg=fe^g=f on UU. \square

Theorem 6.14 (Taking Roots) \quad Let UCU\subset\mathbb C be a simply connected open set, ff a holomorphic function on UU, with f0f\neq 0 pointwise. Then there exists a holomorphic function gg on UU such that gn=fg^n=f.

Proof \quad There exists a holomorphic function hh such that eh=fe^h=f. Set g:=eh/ng:=e^{h/n}, then gn=fg^n=f. \square

Open Mapping Theorem, Biholomorphic Mappings, Local Normal Form#

Definition 6.15 (Biholomorphic function) \quad Let U,VU,V be two open sets in C\mathbb C. A holomorphic function f:UVf:U\to V is called biholomorphic if ff is bijective and f1VUf^{-1}V\to U is holomorphic.

By the inverse function theorem, if f:UCf:U\to\mathbb C is holomorphic, z0Uz_0\in U, and f(z0)0f'(z_0)\neq 0, then ff is locally biholomorphic at z0z_0.

Theorem 6.16 (Local Normal Form of Holomorphic Functions) \quad Let ff be holomorphic on an open set UCU\subset\mathbb C, z0Uz_0\in U, f(m)(z0)0f^{(m)}(z_0)\neq 0 and jm1,f(j)(z0)=0\forall j\leq m-1,\,f^{(j)}(z_0)=0. Then there exist D:=D(z0,r)UD:=D(z_0,r)\subset U and a biholomorphic function hh on DD such that on DD, ff(z0)=hmf-f(z_0)=h^m.

Proof \quad Locally on D:=D(z0,r)D':=D'(z_0,r) we can write ff(z0)=(zz0)mg(z)f-f(z_0)=(z-z_0)^mg(z), where gg is holomorphic, g(z0)=am0g(z_0)=a_m\neq 0. Let h~\tilde h be a holomorphic function on DD' satisfying h~m=g\tilde h^m=g, h:=(zz0)h~h:=(z-z_0)\tilde h, then hh is holomorphic on DD' and ff(z0)=hmf-f(z_0)=h^m. Note that h(z0)=h~(z0)0h'(z_0)=\tilde h(z_0)\neq 0, so by the inverse function theorem, there exists DDD\subset D' such that hh is biholomorphic on DD. \square

Lemma 6.17 \quad gm(z):CC,zzmg_m(z):\mathbb C\to\mathbb C\,,z\mapsto z^m is a proper and open map; further, if z0z\neq 0, then gm1(z)g_m^{-1}(z) contains exactly mm points.

Proof \quad It is easy to see gm(z)g_m(z) is proper. Let z=reiθz=re^{i\theta}, consider the equation wn=zw^n=z. Let w=ρeiψw=\rho e^{i\psi}, then ρm=r\rho^m=r, mψ=θ+2kπm\psi=\theta+2k\pi. The real number ρ=r1/m\rho=r^{1/m} is unique, while ϕ\phi has exactly mm solutions modulo 2π2\pi. This proves gm1(z)g_m^{-1}(z) has mm points. \square

Theorem 6.18 (Open Mapping Theorem) \quad Let ff be a non-constant holomorphic function on a connected open set UCU\subset\mathbb C, then ff is an open map.

Proof \quad It suffices to show that for any z0Uz_0\in U, there exists r>0r>0 such that for any r<rr'<r, D=D(z0,r)D'=D(z_0,r'), f(D)f(D') is open. Indeed, by Theorem 6.14, there exist m1m\geq 1, a biholomorphic function hh, and a disk D=:D(z0,r)D=:D(z_0,r) such that on DD, ff(z0)=hmf-f(z_0)=h^m. By Lemma 6.15, zzmz\mapsto z^m is open, so such DD is as desired. \square

Theorem 6.19 \quad Let ff be a holomorphic function on U\mathbb U. If ff is injective, then f:Uf(U)f:U\to f(U) is biholomorphic.

Proof \quad It suffices to prove the proposition for each connected component of UU. Below assume UU is connected. f(U)f(U) is open. z0U\forall z_0\in U, f(z0)0f'(z_0)\neq 0, otherwise by Theorem 6.14, ff is at least m(m2)m\,(m\geq 2)-to-one around z0z_0, so by the inverse function theorem, f1f^{-1} is locally holomorphic at f(z0)f(z_0). Since z0z_0 is arbitrary, ff is biholomorphic. \square

Maximum Modulus Principle#

Theorem 6.20 (Maximum Modulus Principle) \quad Let UCU\subset\mathbb C be a connected open set, ff a non-constant holomorphic function on UU. Then there does not exist z0Uz_0\in U such that f(z0)=supzUf(z)|f(z_0)|=\sup_{z\in U}|f(z)|.

Proof \quad This is because f(U)f(U) is open. \square

Considering neighborhoods UUU'\subset U for any z0Uz_0\in U in the maximum modulus principle, it further follows that ff cannot attain a local maximum at any z0Uz_0\in U. Similarly, we have the dual “minimum modulus principle”.

Proposition 6.21 \quad Let UCU\subset\mathbb C be a connected open set, ff a non-constant holomorphic function on UU, with f0f\neq 0 pointwise. Then there does not exist z0Uz_0\in U such that f(z0)=infzUf(z)|f(z_0)|=\inf_{z\in U}|f(z)|.

Corollary 6.22 \quad Let UCU\subset\mathbb C be a bounded connected open set, ff a non-constant holomorphic function on U\overline U. Then the maximum of f|f| on U\overline U is attained on U\partial U. If further ff is nonvanishing pointwise, then the minimum of f|f| is attained on U\partial U.

Laurent Series#

Let z0Cz_0\in\mathbb C. A Laurent series centered at z0z_0 is a series of the form:

n=1bn(zz0)n+n=0an(zz0)n,\sum_{n=1}^{\infty}b_n(z-z_0)^{-n}+\sum_{n=0}^{\infty}a_n(z-z_0)^n,

where an,bnCa_n,b_n\in\mathbb C. n=1bn(zz0)n\sum_{n=1}^{\infty}b_n(z-z_0)^{-n} is called the principal part, n=0an(zz0)n\sum_{n=0}^{\infty}a_n(z-z_0)^n the regular part.

Let f:=n=1bn(zz0)n+n=0an(zz0)nf:=\sum_{n=1}^{\infty}b_n(z-z_0)^{-n}+\sum_{n=0}^{\infty}a_n(z-z_0)^n, RR the radius of convergence of its regular part. For the principal part, we can also define a “radius of divergence” rr. Indeed, set S:=j=0bjzjS:=\sum_{j=0}^{\infty}b_jz^j, consider the radius of convergence RR' of SS, then the principal part converges if and only if S((zz0)1)S\big((z-z_0)^{-1}\big) converges. Thus if (zz0)1>R|(z-z_0)^{-1}|>R' i.e., zz0<1/R|z-z_0|<1/R', the principal part does not converge, conversely, if zz0>1/R|z-z_0|>1/R', the principal part converges. So we can set r=1/Rr=1/R'. Define the open annulus

D(z0,r,R):={zC:r<zz0<R},D(z_0,r,R):=\{z\in\mathbb C:r<|z-z_0|<R\},

then ff converges uniformly on compact subsets of D(z0,r,R)D(z_0,r,R). Since each partial sum is holomorphic, ff is holomorphic on D(z0,r,R)D(z_0,r,R).

Lemma 7.1 \quad Let gg be a holomorphic function on D(z0,r,R)D(z_0,r,R), r<r1<r2<Rr<r_1<r_2<R. Then

D(z0,r1)g(z)dz=D(z0,r2)g(z)dz.\int_{\partial D(z_0,r_1)}g(z)\,dz=\int_{\partial D(z_0,r_2)}g(z)\,dz.

Proof \quad Set γ:=D(z0,r1)D(z0,r2)\gamma:=\partial D(z_0,r_1)-\partial D(z_0,r_2). Since D(z0,rj)\partial D(z_0,r_j) are homotopic, γ\gamma is homologous to 00 in D(z0,r,R)D(z_0,r,R), so by the general form of Cauchy’s integral theorem, γgdz=0\int_{\gamma}g\,dz=0, proving the lemma. \square

Theorem 7.2 \quad Let f=n=an(zz0)nf=\sum_{n=-\infty}^{\infty}a_n(z-z_0)^n be a Laurent series convergent on D(z0,r,R)D(z_0,r,R), r<ρ<Rr<\rho<R. Then

an=12πiD(z0,ρ)f(w)(wz0)n+1dw.a_n=\frac{1}{2\pi i}\int_{\partial D(z_0,\rho)}\frac{f(w)}{(w-z_0)^{n+1}}\,dw.

Proof \quad By uniform convergence,

D(z0,ρ)f(w)(wz0)n+1dw=k=D(z0,ρ)akdw(wz0)n+1k.\int_{\partial D(z_0,\rho)}\frac{f(w)}{(w-z_0)^{n+1}}\,dw=\sum_{k=-\infty}^{\infty}\int_{\partial D(z_0,\rho)}\frac{a_k\,dw}{(w-z_0)^{n+1-k}}.

If knk\neq n then ak(wz0)kn1a_k(w-z_0)^{k-n-1} is exact, its integral is zero. So

D(z0,ρ)f(w)(wz0)n+1dw=D(z0,ρ)andwwz0=2πian.\begin{aligned} \int_{\partial D(z_0,\rho)}\frac{f(w)}{(w-z_0)^{n+1}}\,dw=\int_{\partial D(z_0,\rho)}\frac{a_n\,dw}{w-z_0}=2\pi i\cdot a_n. \end{aligned}

\square

Theorem 7.3 (Laurent Expansion) \quad Let ff be a holomorphic function on D(z0,r,R)D(z_0,r,R), then ff can be represented as a Laurent series centered at z0z_0 on D(z0,r,R)D(z_0,r,R).

Proof \quad By Theorem 7.2, the coefficients of the Laurent series, if they exist, are unique. It suffices to show that ff can be expanded as a Laurent series on each D(z0,r,R),r<r<R<RD(z_0,r',R'),\,r<r'<R'<R. For each zD(z0,r,R)z\in D(z_0,r',R'), n(D(z0,R)D(z0,r),z)1n(\partial D(z_0,R')-\partial D(z_0,r'),z)\equiv 1. Then by the general form of Cauchy’s integral theorem,

f(z)=12πiD(z0,R)f(w)wzdw12πiD(z0,r)f(w)wzdw,f(z)=\frac{1}{2\pi i}\int_{\partial D(z_0,R')}\frac{f(w)}{w-z}\,dw-\frac{1}{2\pi i}\int_{\partial D(z_0,r')}\frac{f(w)}{w-z}\,dw,

for fixed zD(z0,r,R)z\in D(z_0,r',R'), consider the series

wz0wz=n=0(zz0wz0)n,\frac{w-z_0}{w-z}=\sum_{n=0}^{\infty}\left(\frac{z-z_0}{w-z_0}\right)^n,

it converges for wD(z0,R)w\in D(z_0,R'), so

D(z0,R)f(w)wzdw=n=0(D(z0,R)f(w)dw(wz0)n+1)(zz0)n.\int_{\partial D(z_0,R')}\frac{f(w)}{w-z}\,dw=\sum_{n=0}^{\infty}\left(\int_{\partial D(z_0,R')}\frac{f(w)\,dw}{(w-z_0)^{n+1}}\right)(z-z_0)^n.

For fixed zD(z0,r,R)z\in D(z_0,r',R'), consider the series

zz0zw=n=0(wz0zz0)n,\frac{z-z_0}{z-w}=\sum_{n=0}^{\infty}\left(\frac{w-z_0}{z-z_0}\right)^n,

it converges for wD(z0,r)w\in D(z_0,r'), so

D(z0,r)f(w)wzdw=n=0(D(z0,r)(wz0)nf(w)dw)(zz0)n1=n=1(D(z0,r)f(w)dw(wz0)n+1)(zz0)n.\begin{aligned} \int_{\partial D(z_0,r')}\frac{f(w)}{w-z}\,dw&=-\sum_{n=0}^{\infty}\left(\int_{\partial D(z_0,r')}(w-z_0)^nf(w)\,dw\right)(z-z_0)^{-n-1}\\ &=-\sum_{n=-1}^{-\infty}\left(\int_{\partial D(z_0,r')}\frac{f(w)\,dw}{(w-z_0)^{n+1}}\right)(z-z_0)^n. \end{aligned}

Combining these two parts proves the proposition. \square

From Theorem 7.2 and Theorem 7.3, if ff is holomorphic on D(z0,r,R)D(z_0,r,R), then ff can be expanded as

f(z)=12πin=(D(z0,ρ)f(ζ)dζ(ζz0)n+1)(zz0)n.f(z)=\frac{1}{2\pi i}\sum_{n=-\infty}^{\infty}\left(\int_{\partial D(z_0,\rho)}\frac{f(\zeta)\,d\zeta}{(\zeta-z_0)^{n+1}}\right)(z-z_0)^n.
Analytic Functions and Power Series Theory
https://peakyi.github.io/posts/analytic-functions-and-power-series-theory/
Author
Anby Demara
Published at
2025-12-10
License
CC BY-NC-SA 4.0